2 * Copyright (C) 2013 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 * SPDX-License-Identifier: GPL-2.0-only
22 #include <sys/mount.h>
23 #include <sys/resource.h>
24 #include <sys/socket.h>
26 #include <sys/types.h>
29 #include <urcu/futex.h>
30 #include <urcu/rculist.h>
31 #include <urcu/uatomic.h>
33 #include <common/common.h>
34 #include <common/compat/endian.h>
35 #include <common/compat/poll.h>
36 #include <common/compat/socket.h>
37 #include <common/defaults.h>
38 #include <common/fd-tracker/utils.h>
39 #include <common/fs-handle.h>
40 #include <common/futex.h>
41 #include <common/index/index.h>
42 #include <common/sessiond-comm/inet.h>
43 #include <common/sessiond-comm/relayd.h>
44 #include <common/sessiond-comm/sessiond-comm.h>
45 #include <common/uri.h>
46 #include <common/utils.h>
47 #include <lttng/lttng.h>
50 #include "connection.h"
51 #include "ctf-trace.h"
52 #include "health-relayd.h"
54 #include "lttng-relayd.h"
57 #include "testpoint.h"
59 #include "viewer-session.h"
60 #include "viewer-stream.h"
62 #define SESSION_BUF_DEFAULT_COUNT 16
64 static struct lttng_uri
*live_uri
;
67 * This pipe is used to inform the worker thread that a command is queued and
68 * ready to be processed.
70 static int live_conn_pipe
[2] = { -1, -1 };
72 /* Shared between threads */
73 static int live_dispatch_thread_exit
;
75 static pthread_t live_listener_thread
;
76 static pthread_t live_dispatcher_thread
;
77 static pthread_t live_worker_thread
;
80 * Relay command queue.
82 * The live_thread_listener and live_thread_dispatcher communicate with this
85 static struct relay_conn_queue viewer_conn_queue
;
87 static uint64_t last_relay_viewer_session_id
;
88 static pthread_mutex_t last_relay_viewer_session_id_lock
=
89 PTHREAD_MUTEX_INITIALIZER
;
95 void cleanup_relayd_live(void)
103 * Receive a request buffer using a given socket, destination allocated buffer
106 * Return the size of the received message or else a negative value on error
107 * with errno being set by recvmsg() syscall.
110 ssize_t
recv_request(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
114 ret
= sock
->ops
->recvmsg(sock
, buf
, size
, 0);
115 if (ret
< 0 || ret
!= size
) {
117 /* Orderly shutdown. Not necessary to print an error. */
118 DBG("Socket %d did an orderly shutdown", sock
->fd
);
120 ERR("Relay failed to receive request.");
129 * Send a response buffer using a given socket, source allocated buffer of
132 * Return the size of the sent message or else a negative value on error with
133 * errno being set by sendmsg() syscall.
136 ssize_t
send_response(struct lttcomm_sock
*sock
, void *buf
, size_t size
)
140 ret
= sock
->ops
->sendmsg(sock
, buf
, size
, 0);
142 ERR("Relayd failed to send response.");
149 * Atomically check if new streams got added in one of the sessions attached
150 * and reset the flag to 0.
152 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
156 int check_new_streams(struct relay_connection
*conn
)
158 struct relay_session
*session
;
159 unsigned long current_val
;
162 if (!conn
->viewer_session
) {
166 cds_list_for_each_entry_rcu(session
,
167 &conn
->viewer_session
->session_list
,
168 viewer_session_node
) {
169 if (!session_get(session
)) {
172 current_val
= uatomic_cmpxchg(&session
->new_streams
, 1, 0);
174 session_put(session
);
185 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
186 * this function should ignore the sent flag or not.
188 * Return 0 on success or else a negative value.
191 ssize_t
send_viewer_streams(struct lttcomm_sock
*sock
,
192 uint64_t session_id
, unsigned int ignore_sent_flag
)
195 struct lttng_ht_iter iter
;
196 struct relay_viewer_stream
*vstream
;
200 cds_lfht_for_each_entry(viewer_streams_ht
->ht
, &iter
.iter
, vstream
,
202 struct ctf_trace
*ctf_trace
;
203 struct lttng_viewer_stream send_stream
= {};
205 health_code_update();
207 if (!viewer_stream_get(vstream
)) {
211 pthread_mutex_lock(&vstream
->stream
->lock
);
212 /* Ignore if not the same session. */
213 if (vstream
->stream
->trace
->session
->id
!= session_id
||
214 (!ignore_sent_flag
&& vstream
->sent_flag
)) {
215 pthread_mutex_unlock(&vstream
->stream
->lock
);
216 viewer_stream_put(vstream
);
220 ctf_trace
= vstream
->stream
->trace
;
221 send_stream
.id
= htobe64(vstream
->stream
->stream_handle
);
222 send_stream
.ctf_trace_id
= htobe64(ctf_trace
->id
);
223 send_stream
.metadata_flag
= htobe32(
224 vstream
->stream
->is_metadata
);
225 if (lttng_strncpy(send_stream
.path_name
, vstream
->path_name
,
226 sizeof(send_stream
.path_name
))) {
227 pthread_mutex_unlock(&vstream
->stream
->lock
);
228 viewer_stream_put(vstream
);
229 ret
= -1; /* Error. */
232 if (lttng_strncpy(send_stream
.channel_name
,
233 vstream
->channel_name
,
234 sizeof(send_stream
.channel_name
))) {
235 pthread_mutex_unlock(&vstream
->stream
->lock
);
236 viewer_stream_put(vstream
);
237 ret
= -1; /* Error. */
241 DBG("Sending stream %" PRIu64
" to viewer",
242 vstream
->stream
->stream_handle
);
243 vstream
->sent_flag
= 1;
244 pthread_mutex_unlock(&vstream
->stream
->lock
);
246 ret
= send_response(sock
, &send_stream
, sizeof(send_stream
));
247 viewer_stream_put(vstream
);
261 * Create every viewer stream possible for the given session with the seek
262 * type. Three counters *can* be return which are in order the total amount of
263 * viewer stream of the session, the number of unsent stream and the number of
264 * stream created. Those counters can be NULL and thus will be ignored.
266 * session must be locked to ensure that we see either none or all initial
267 * streams for a session, but no intermediate state..
269 * Return 0 on success or else a negative value.
271 static int make_viewer_streams(struct relay_session
*relay_session
,
272 struct relay_viewer_session
*viewer_session
,
273 enum lttng_viewer_seek seek_t
,
276 uint32_t *nb_created
,
280 struct lttng_ht_iter iter
;
281 struct ctf_trace
*ctf_trace
;
282 struct relay_stream
*relay_stream
= NULL
;
284 LTTNG_ASSERT(relay_session
);
285 ASSERT_LOCKED(relay_session
->lock
);
287 if (relay_session
->connection_closed
) {
292 * Create viewer streams for relay streams that are ready to be
293 * used for a the given session id only.
296 cds_lfht_for_each_entry (relay_session
->ctf_traces_ht
->ht
, &iter
.iter
,
297 ctf_trace
, node
.node
) {
298 bool trace_has_metadata_stream
= false;
300 health_code_update();
302 if (!ctf_trace_get(ctf_trace
)) {
307 * Iterate over all the streams of the trace to see if we have a
310 cds_list_for_each_entry_rcu(relay_stream
,
311 &ctf_trace
->stream_list
, stream_node
)
313 bool is_metadata_stream
;
315 pthread_mutex_lock(&relay_stream
->lock
);
316 is_metadata_stream
= relay_stream
->is_metadata
;
317 pthread_mutex_unlock(&relay_stream
->lock
);
319 if (is_metadata_stream
) {
320 trace_has_metadata_stream
= true;
328 * If there is no metadata stream in this trace at the moment
329 * and we never sent one to the viewer, skip the trace. We
330 * accept that the viewer will not see this trace at all.
332 if (!trace_has_metadata_stream
&&
333 !ctf_trace
->metadata_stream_sent_to_viewer
) {
334 ctf_trace_put(ctf_trace
);
338 cds_list_for_each_entry_rcu(relay_stream
,
339 &ctf_trace
->stream_list
, stream_node
)
341 struct relay_viewer_stream
*viewer_stream
;
343 if (!stream_get(relay_stream
)) {
347 pthread_mutex_lock(&relay_stream
->lock
);
349 * stream published is protected by the session lock.
351 if (!relay_stream
->published
) {
354 viewer_stream
= viewer_stream_get_by_id(
355 relay_stream
->stream_handle
);
356 if (!viewer_stream
) {
357 struct lttng_trace_chunk
*viewer_stream_trace_chunk
= NULL
;
360 * Save that we sent the metadata stream to the
361 * viewer. So that we know what trace the viewer
364 if (relay_stream
->is_metadata
) {
365 ctf_trace
->metadata_stream_sent_to_viewer
= true;
369 * If a rotation is ongoing, use a copy of the
370 * relay stream's chunk to ensure the stream
373 * Otherwise, the viewer session's current trace
374 * chunk can be used safely.
376 if ((relay_stream
->ongoing_rotation
.is_set
||
377 relay_session
->ongoing_rotation
) &&
378 relay_stream
->trace_chunk
) {
379 viewer_stream_trace_chunk
= lttng_trace_chunk_copy(
380 relay_stream
->trace_chunk
);
381 if (!viewer_stream_trace_chunk
) {
383 ctf_trace_put(ctf_trace
);
388 * Transition the viewer session into the newest trace chunk available.
390 if (!lttng_trace_chunk_ids_equal(viewer_session
->current_trace_chunk
,
391 relay_stream
->trace_chunk
)) {
393 ret
= viewer_session_set_trace_chunk_copy(
395 relay_stream
->trace_chunk
);
398 ctf_trace_put(ctf_trace
);
403 if (relay_stream
->trace_chunk
) {
405 * If the corresponding relay
406 * stream's trace chunk is set,
407 * the viewer stream will be
410 * Note that a relay stream can
411 * have a NULL output trace
412 * chunk (for instance, after a
413 * clear against a stopped
416 const bool reference_acquired
= lttng_trace_chunk_get(
417 viewer_session
->current_trace_chunk
);
419 LTTNG_ASSERT(reference_acquired
);
420 viewer_stream_trace_chunk
=
421 viewer_session
->current_trace_chunk
;
425 viewer_stream
= viewer_stream_create(
427 viewer_stream_trace_chunk
,
429 lttng_trace_chunk_put(viewer_stream_trace_chunk
);
430 viewer_stream_trace_chunk
= NULL
;
431 if (!viewer_stream
) {
433 ctf_trace_put(ctf_trace
);
438 /* Update number of created stream counter. */
442 * Ensure a self-reference is preserved even
443 * after we have put our local reference.
445 if (!viewer_stream_get(viewer_stream
)) {
446 ERR("Unable to get self-reference on viewer stream, logic error.");
450 if (!viewer_stream
->sent_flag
&& nb_unsent
) {
451 /* Update number of unsent stream counter. */
455 /* Update number of total stream counter. */
457 if (relay_stream
->is_metadata
) {
458 if (!relay_stream
->closed
||
459 relay_stream
->metadata_received
>
460 viewer_stream
->metadata_sent
) {
464 if (!relay_stream
->closed
||
465 !(((int64_t)(relay_stream
->prev_data_seq
-
466 relay_stream
->last_net_seq_num
)) >=
472 /* Put local reference. */
473 viewer_stream_put(viewer_stream
);
475 pthread_mutex_unlock(&relay_stream
->lock
);
476 stream_put(relay_stream
);
479 ctf_trace_put(ctf_trace
);
488 pthread_mutex_unlock(&relay_stream
->lock
);
489 stream_put(relay_stream
);
495 int relayd_live_stop(void)
497 /* Stop dispatch thread */
498 CMM_STORE_SHARED(live_dispatch_thread_exit
, 1);
499 futex_nto1_wake(&viewer_conn_queue
.futex
);
504 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
507 int create_named_thread_poll_set(struct lttng_poll_event
*events
,
508 int size
, const char *name
)
512 if (events
== NULL
|| size
== 0) {
517 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
518 name
, events
, 1, LTTNG_CLOEXEC
);
520 PERROR("Failed to create \"%s\" poll file descriptor", name
);
525 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
537 * Check if the thread quit pipe was triggered.
539 * Return 1 if it was triggered else 0;
542 int check_thread_quit_pipe(int fd
, uint32_t events
)
544 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
552 int create_sock(void *data
, int *out_fd
)
555 struct lttcomm_sock
*sock
= (lttcomm_sock
*) data
;
557 ret
= lttcomm_create_sock(sock
);
568 int close_sock(void *data
, int *in_fd
)
570 struct lttcomm_sock
*sock
= (lttcomm_sock
*) data
;
572 return sock
->ops
->close(sock
);
575 static int accept_sock(void *data
, int *out_fd
)
578 /* Socks is an array of in_sock, out_sock. */
579 struct lttcomm_sock
**socks
= (lttcomm_sock
**) data
;
580 struct lttcomm_sock
*in_sock
= socks
[0];
582 socks
[1] = in_sock
->ops
->accept(in_sock
);
587 *out_fd
= socks
[1]->fd
;
593 struct lttcomm_sock
*accept_live_sock(struct lttcomm_sock
*listening_sock
,
597 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
598 struct lttcomm_sock
*new_sock
= NULL
;
600 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &out_fd
,
601 (const char **) &name
, 1, accept_sock
, &socks
);
606 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
612 * Create and init socket from uri.
615 struct lttcomm_sock
*init_socket(struct lttng_uri
*uri
, const char *name
)
618 struct lttcomm_sock
*sock
= NULL
;
619 char uri_str
[LTTNG_PATH_MAX
];
620 char *formated_name
= NULL
;
622 sock
= lttcomm_alloc_sock_from_uri(uri
);
624 ERR("Allocating socket");
629 * Don't fail to create the socket if the name can't be built as it is
630 * only used for debugging purposes.
632 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
633 uri_str
[sizeof(uri_str
) - 1] = '\0';
635 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
638 formated_name
= NULL
;
642 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
643 (const char **) (formated_name
? &formated_name
: NULL
),
644 1, create_sock
, sock
);
646 PERROR("Failed to create \"%s\" socket",
647 formated_name
?: "Unknown");
650 DBG("Listening on %s socket %d", name
, sock
->fd
);
652 ret
= sock
->ops
->bind(sock
);
654 PERROR("Failed to bind lttng-live socket");
658 ret
= sock
->ops
->listen(sock
, -1);
669 lttcomm_destroy_sock(sock
);
676 * This thread manages the listening for new connections on the network
679 void *thread_listener(void *data
)
681 int i
, ret
, pollfd
, err
= -1;
682 uint32_t revents
, nb_fd
;
683 struct lttng_poll_event events
;
684 struct lttcomm_sock
*live_control_sock
;
686 DBG("[thread] Relay live listener started");
688 rcu_register_thread();
689 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_LISTENER
);
691 health_code_update();
693 live_control_sock
= init_socket(live_uri
, "Live listener");
694 if (!live_control_sock
) {
695 goto error_sock_control
;
698 /* Pass 2 as size here for the thread quit pipe and control sockets. */
699 ret
= create_named_thread_poll_set(&events
, 2,
700 "Live listener thread epoll");
702 goto error_create_poll
;
705 /* Add the control socket */
706 ret
= lttng_poll_add(&events
, live_control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
711 lttng_relay_notify_ready();
713 if (testpoint(relayd_thread_live_listener
)) {
714 goto error_testpoint
;
718 health_code_update();
720 DBG("Listener accepting live viewers connections");
724 ret
= lttng_poll_wait(&events
, -1);
728 * Restart interrupted system call.
730 if (errno
== EINTR
) {
737 DBG("Relay new viewer connection received");
738 for (i
= 0; i
< nb_fd
; i
++) {
739 health_code_update();
741 /* Fetch once the poll data */
742 revents
= LTTNG_POLL_GETEV(&events
, i
);
743 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
745 /* Thread quit pipe has been closed. Killing thread. */
746 ret
= check_thread_quit_pipe(pollfd
, revents
);
752 if (revents
& LPOLLIN
) {
754 * A new connection is requested, therefore a
755 * viewer connection is allocated in this
756 * thread, enqueued to a global queue and
757 * dequeued (and freed) in the worker thread.
760 struct relay_connection
*new_conn
;
761 struct lttcomm_sock
*newsock
;
763 newsock
= accept_live_sock(live_control_sock
,
764 "Live socket to client");
766 PERROR("accepting control sock");
769 DBG("Relay viewer connection accepted socket %d", newsock
->fd
);
771 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
774 PERROR("setsockopt inet");
775 lttcomm_destroy_sock(newsock
);
778 new_conn
= connection_create(newsock
, RELAY_CONNECTION_UNKNOWN
);
780 lttcomm_destroy_sock(newsock
);
783 /* Ownership assumed by the connection. */
786 /* Enqueue request for the dispatcher thread. */
787 cds_wfcq_head_ptr_t head
;
788 head
.h
= &viewer_conn_queue
.head
;
789 cds_wfcq_enqueue(head
, &viewer_conn_queue
.tail
,
793 * Wake the dispatch queue futex.
794 * Implicit memory barrier with the
795 * exchange in cds_wfcq_enqueue.
797 futex_nto1_wake(&viewer_conn_queue
.futex
);
798 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
799 ERR("socket poll error");
802 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
812 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
814 if (live_control_sock
->fd
>= 0) {
815 int sock_fd
= live_control_sock
->fd
;
817 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
818 &sock_fd
, 1, close_sock
,
823 live_control_sock
->fd
= -1;
825 lttcomm_destroy_sock(live_control_sock
);
829 DBG("Live viewer listener thread exited with error");
831 health_unregister(health_relayd
);
832 rcu_unregister_thread();
833 DBG("Live viewer listener thread cleanup complete");
834 if (lttng_relay_stop_threads()) {
835 ERR("Error stopping threads");
841 * This thread manages the dispatching of the requests to worker threads
844 void *thread_dispatcher(void *data
)
848 struct cds_wfcq_node
*node
;
849 struct relay_connection
*conn
= NULL
;
851 DBG("[thread] Live viewer relay dispatcher started");
853 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER
);
855 if (testpoint(relayd_thread_live_dispatcher
)) {
856 goto error_testpoint
;
859 health_code_update();
862 health_code_update();
864 /* Atomically prepare the queue futex */
865 futex_nto1_prepare(&viewer_conn_queue
.futex
);
867 if (CMM_LOAD_SHARED(live_dispatch_thread_exit
)) {
872 health_code_update();
874 /* Dequeue commands */
875 node
= cds_wfcq_dequeue_blocking(&viewer_conn_queue
.head
,
876 &viewer_conn_queue
.tail
);
878 DBG("Woken up but nothing in the live-viewer "
879 "relay command queue");
880 /* Continue thread execution */
883 conn
= caa_container_of(node
, struct relay_connection
, qnode
);
884 DBG("Dispatching viewer request waiting on sock %d",
888 * Inform worker thread of the new request. This
889 * call is blocking so we can be assured that
890 * the data will be read at some point in time
891 * or wait to the end of the world :)
893 ret
= lttng_write(live_conn_pipe
[1], &conn
, sizeof(conn
));
895 PERROR("write conn pipe");
896 connection_put(conn
);
899 } while (node
!= NULL
);
901 /* Futex wait on queue. Blocking call on futex() */
903 futex_nto1_wait(&viewer_conn_queue
.futex
);
907 /* Normal exit, no error */
914 ERR("Health error occurred in %s", __func__
);
916 health_unregister(health_relayd
);
917 DBG("Live viewer dispatch thread dying");
918 if (lttng_relay_stop_threads()) {
919 ERR("Error stopping threads");
925 * Establish connection with the viewer and check the versions.
927 * Return 0 on success or else negative value.
930 int viewer_connect(struct relay_connection
*conn
)
933 struct lttng_viewer_connect reply
, msg
;
935 conn
->version_check_done
= 1;
937 health_code_update();
939 DBG("Viewer is establishing a connection to the relayd.");
941 ret
= recv_request(conn
->sock
, &msg
, sizeof(msg
));
946 health_code_update();
948 memset(&reply
, 0, sizeof(reply
));
949 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
950 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
952 /* Major versions must be the same */
953 if (reply
.major
!= be32toh(msg
.major
)) {
954 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
955 reply
.major
, be32toh(msg
.major
));
960 conn
->major
= reply
.major
;
961 /* We adapt to the lowest compatible version */
962 if (reply
.minor
<= be32toh(msg
.minor
)) {
963 conn
->minor
= reply
.minor
;
965 conn
->minor
= be32toh(msg
.minor
);
968 if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_COMMAND
) {
969 conn
->type
= RELAY_VIEWER_COMMAND
;
970 } else if (be32toh(msg
.type
) == LTTNG_VIEWER_CLIENT_NOTIFICATION
) {
971 conn
->type
= RELAY_VIEWER_NOTIFICATION
;
973 ERR("Unknown connection type : %u", be32toh(msg
.type
));
978 reply
.major
= htobe32(reply
.major
);
979 reply
.minor
= htobe32(reply
.minor
);
980 if (conn
->type
== RELAY_VIEWER_COMMAND
) {
982 * Increment outside of htobe64 macro, because the argument can
983 * be used more than once within the macro, and thus the
984 * operation may be undefined.
986 pthread_mutex_lock(&last_relay_viewer_session_id_lock
);
987 last_relay_viewer_session_id
++;
988 pthread_mutex_unlock(&last_relay_viewer_session_id_lock
);
989 reply
.viewer_session_id
= htobe64(last_relay_viewer_session_id
);
992 health_code_update();
994 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
999 health_code_update();
1001 DBG("Version check done using protocol %u.%u", conn
->major
, conn
->minor
);
1009 * Send the viewer the list of current sessions.
1010 * We need to create a copy of the hash table content because otherwise
1011 * we cannot assume the number of entries stays the same between getting
1012 * the number of HT elements and iteration over the HT.
1014 * Return 0 on success or else a negative value.
1017 int viewer_list_sessions(struct relay_connection
*conn
)
1020 struct lttng_viewer_list_sessions session_list
;
1021 struct lttng_ht_iter iter
;
1022 struct relay_session
*session
;
1023 struct lttng_viewer_session
*send_session_buf
= NULL
;
1024 uint32_t buf_count
= SESSION_BUF_DEFAULT_COUNT
;
1027 DBG("List sessions received");
1029 send_session_buf
= (lttng_viewer_session
*) zmalloc(SESSION_BUF_DEFAULT_COUNT
* sizeof(*send_session_buf
));
1030 if (!send_session_buf
) {
1035 cds_lfht_for_each_entry(sessions_ht
->ht
, &iter
.iter
, session
,
1037 struct lttng_viewer_session
*send_session
;
1039 health_code_update();
1041 pthread_mutex_lock(&session
->lock
);
1042 if (session
->connection_closed
) {
1043 /* Skip closed session */
1047 if (count
>= buf_count
) {
1048 struct lttng_viewer_session
*newbuf
;
1049 uint32_t new_buf_count
= buf_count
<< 1;
1051 newbuf
= (lttng_viewer_session
*) realloc(send_session_buf
,
1052 new_buf_count
* sizeof(*send_session_buf
));
1057 send_session_buf
= newbuf
;
1058 buf_count
= new_buf_count
;
1060 send_session
= &send_session_buf
[count
];
1061 if (lttng_strncpy(send_session
->session_name
,
1062 session
->session_name
,
1063 sizeof(send_session
->session_name
))) {
1067 if (lttng_strncpy(send_session
->hostname
, session
->hostname
,
1068 sizeof(send_session
->hostname
))) {
1072 send_session
->id
= htobe64(session
->id
);
1073 send_session
->live_timer
= htobe32(session
->live_timer
);
1074 if (session
->viewer_attached
) {
1075 send_session
->clients
= htobe32(1);
1077 send_session
->clients
= htobe32(0);
1079 send_session
->streams
= htobe32(session
->stream_count
);
1082 pthread_mutex_unlock(&session
->lock
);
1085 pthread_mutex_unlock(&session
->lock
);
1093 session_list
.sessions_count
= htobe32(count
);
1095 health_code_update();
1097 ret
= send_response(conn
->sock
, &session_list
, sizeof(session_list
));
1102 health_code_update();
1104 ret
= send_response(conn
->sock
, send_session_buf
,
1105 count
* sizeof(*send_session_buf
));
1109 health_code_update();
1113 free(send_session_buf
);
1118 * Send the viewer the list of current streams.
1121 int viewer_get_new_streams(struct relay_connection
*conn
)
1123 int ret
, send_streams
= 0;
1124 uint32_t nb_created
= 0, nb_unsent
= 0, nb_streams
= 0, nb_total
= 0;
1125 struct lttng_viewer_new_streams_request request
;
1126 struct lttng_viewer_new_streams_response response
;
1127 struct relay_session
*session
= NULL
;
1128 uint64_t session_id
;
1129 bool closed
= false;
1133 DBG("Get new streams received");
1135 health_code_update();
1137 /* Receive the request from the connected client. */
1138 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1142 session_id
= be64toh(request
.session_id
);
1144 health_code_update();
1146 memset(&response
, 0, sizeof(response
));
1148 session
= session_get_by_id(session_id
);
1150 DBG("Relay session %" PRIu64
" not found", session_id
);
1151 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
1155 if (!viewer_session_is_attached(conn
->viewer_session
, session
)) {
1156 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR
);
1161 * For any new stream, create it with LTTNG_VIEWER_SEEK_BEGINNING since
1162 * that at this point the client is already attached to the session.Aany
1163 * initial stream will have been created with the seek type at attach
1164 * time (for now most readers use the LTTNG_VIEWER_SEEK_LAST on attach).
1165 * Otherwise any event happening in a new stream between the attach and
1166 * a call to viewer_get_new_streams will be "lost" (never received) from
1167 * the viewer's point of view.
1169 pthread_mutex_lock(&session
->lock
);
1170 ret
= make_viewer_streams(session
,
1171 conn
->viewer_session
,
1172 LTTNG_VIEWER_SEEK_BEGINNING
, &nb_total
, &nb_unsent
,
1173 &nb_created
, &closed
);
1175 goto error_unlock_session
;
1178 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_OK
);
1180 /* Only send back the newly created streams with the unsent ones. */
1181 nb_streams
= nb_created
+ nb_unsent
;
1182 response
.streams_count
= htobe32(nb_streams
);
1185 * If the session is closed, HUP when there are no more streams
1188 if (closed
&& nb_total
== 0) {
1190 response
.streams_count
= 0;
1191 response
.status
= htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP
);
1192 goto send_reply_unlock
;
1195 pthread_mutex_unlock(&session
->lock
);
1198 health_code_update();
1199 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
1201 goto end_put_session
;
1203 health_code_update();
1206 * Unknown or empty session, just return gracefully, the viewer
1207 * knows what is happening.
1209 if (!send_streams
|| !nb_streams
) {
1211 goto end_put_session
;
1215 * Send stream and *DON'T* ignore the sent flag so every viewer
1216 * streams that were not sent from that point will be sent to
1219 ret
= send_viewer_streams(conn
->sock
, session_id
, 0);
1221 goto end_put_session
;
1226 session_put(session
);
1230 error_unlock_session
:
1231 pthread_mutex_unlock(&session
->lock
);
1232 session_put(session
);
1237 * Send the viewer the list of current sessions.
1240 int viewer_attach_session(struct relay_connection
*conn
)
1242 int send_streams
= 0;
1244 uint32_t nb_streams
= 0;
1245 enum lttng_viewer_seek seek_type
;
1246 struct lttng_viewer_attach_session_request request
;
1247 struct lttng_viewer_attach_session_response response
;
1248 struct relay_session
*session
= NULL
;
1249 enum lttng_viewer_attach_return_code viewer_attach_status
;
1250 bool closed
= false;
1251 uint64_t session_id
;
1255 health_code_update();
1257 /* Receive the request from the connected client. */
1258 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1263 session_id
= be64toh(request
.session_id
);
1264 health_code_update();
1266 memset(&response
, 0, sizeof(response
));
1268 if (!conn
->viewer_session
) {
1269 DBG("Client trying to attach before creating a live viewer session");
1270 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION
);
1274 session
= session_get_by_id(session_id
);
1276 DBG("Relay session %" PRIu64
" not found", session_id
);
1277 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
1280 DBG("Attach session ID %" PRIu64
" received", session_id
);
1282 pthread_mutex_lock(&session
->lock
);
1283 if (session
->live_timer
== 0) {
1284 DBG("Not live session");
1285 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE
);
1290 viewer_attach_status
= viewer_session_attach(conn
->viewer_session
,
1292 if (viewer_attach_status
!= LTTNG_VIEWER_ATTACH_OK
) {
1293 response
.status
= htobe32(viewer_attach_status
);
1297 switch (be32toh(request
.seek
)) {
1298 case LTTNG_VIEWER_SEEK_BEGINNING
:
1299 case LTTNG_VIEWER_SEEK_LAST
:
1300 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_OK
);
1301 seek_type
= (lttng_viewer_seek
) be32toh(request
.seek
);
1304 ERR("Wrong seek parameter");
1305 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR
);
1310 ret
= make_viewer_streams(session
,
1311 conn
->viewer_session
, seek_type
,
1312 &nb_streams
, NULL
, NULL
, &closed
);
1314 goto end_put_session
;
1316 pthread_mutex_unlock(&session
->lock
);
1317 session_put(session
);
1320 response
.streams_count
= htobe32(nb_streams
);
1322 * If the session is closed when the viewer is attaching, it
1323 * means some of the streams may have been concurrently removed,
1324 * so we don't allow the viewer to attach, even if there are
1325 * streams available.
1329 response
.streams_count
= 0;
1330 response
.status
= htobe32(LTTNG_VIEWER_ATTACH_UNK
);
1335 health_code_update();
1336 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
1338 goto end_put_session
;
1340 health_code_update();
1343 * Unknown or empty session, just return gracefully, the viewer
1344 * knows what is happening.
1346 if (!send_streams
|| !nb_streams
) {
1348 goto end_put_session
;
1351 /* Send stream and ignore the sent flag. */
1352 ret
= send_viewer_streams(conn
->sock
, session_id
, 1);
1354 goto end_put_session
;
1359 pthread_mutex_unlock(&session
->lock
);
1360 session_put(session
);
1367 * Open the index file if needed for the given vstream.
1369 * If an index file is successfully opened, the vstream will set it as its
1370 * current index file.
1372 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1374 * Called with rstream lock held.
1376 static int try_open_index(struct relay_viewer_stream
*vstream
,
1377 struct relay_stream
*rstream
)
1380 const uint32_t connection_major
= rstream
->trace
->session
->major
;
1381 const uint32_t connection_minor
= rstream
->trace
->session
->minor
;
1382 enum lttng_trace_chunk_status chunk_status
;
1384 if (vstream
->index_file
) {
1389 * First time, we open the index file and at least one index is ready.
1391 if (rstream
->index_received_seqcount
== 0 ||
1392 !vstream
->stream_file
.trace_chunk
) {
1397 chunk_status
= lttng_index_file_create_from_trace_chunk_read_only(
1398 vstream
->stream_file
.trace_chunk
, rstream
->path_name
,
1399 rstream
->channel_name
, rstream
->tracefile_size
,
1400 vstream
->current_tracefile_id
,
1401 lttng_to_index_major(connection_major
, connection_minor
),
1402 lttng_to_index_minor(connection_major
, connection_minor
),
1403 true, &vstream
->index_file
);
1404 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1405 if (chunk_status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
) {
1417 * Check the status of the index for the given stream. This function
1418 * updates the index structure if needed and can put (close) the vstream
1419 * in the HUP situation.
1421 * Return 0 means that we can proceed with the index. A value of 1 means
1422 * that the index has been updated and is ready to be sent to the
1423 * client. A negative value indicates an error that can't be handled.
1425 * Called with rstream lock held.
1427 static int check_index_status(struct relay_viewer_stream
*vstream
,
1428 struct relay_stream
*rstream
, struct ctf_trace
*trace
,
1429 struct lttng_viewer_index
*index
)
1433 DBG("Check index status: index_received_seqcount %" PRIu64
" "
1434 "index_sent_seqcount %" PRIu64
" "
1435 "for stream %" PRIu64
,
1436 rstream
->index_received_seqcount
,
1437 vstream
->index_sent_seqcount
,
1438 vstream
->stream
->stream_handle
);
1439 if ((trace
->session
->connection_closed
|| rstream
->closed
)
1440 && rstream
->index_received_seqcount
1441 == vstream
->index_sent_seqcount
) {
1443 * Last index sent and session connection or relay
1444 * stream are closed.
1446 index
->status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1448 } else if (rstream
->beacon_ts_end
!= -1ULL &&
1449 (rstream
->index_received_seqcount
== 0 ||
1450 (vstream
->index_sent_seqcount
!= 0 &&
1451 rstream
->index_received_seqcount
1452 <= vstream
->index_sent_seqcount
))) {
1454 * We've received a synchronization beacon and the last index
1455 * available has been sent, the index for now is inactive.
1457 * In this case, we have received a beacon which allows us to
1458 * inform the client of a time interval during which we can
1459 * guarantee that there are no events to read (and never will
1462 * The sent seqcount can grow higher than receive seqcount on
1463 * clear because the rotation performed by clear will push
1464 * the index_sent_seqcount ahead (see
1465 * viewer_stream_sync_tracefile_array_tail) and skip over
1466 * packet sequence numbers.
1468 index
->status
= htobe32(LTTNG_VIEWER_INDEX_INACTIVE
);
1469 index
->timestamp_end
= htobe64(rstream
->beacon_ts_end
);
1470 index
->stream_id
= htobe64(rstream
->ctf_stream_id
);
1471 DBG("Check index status: inactive with beacon, for stream %" PRIu64
,
1472 vstream
->stream
->stream_handle
);
1474 } else if (rstream
->index_received_seqcount
== 0 ||
1475 (vstream
->index_sent_seqcount
!= 0 &&
1476 rstream
->index_received_seqcount
1477 <= vstream
->index_sent_seqcount
)) {
1479 * This checks whether received <= sent seqcount. In
1480 * this case, we have not received a beacon. Therefore,
1481 * we can only ask the client to retry later.
1483 * The sent seqcount can grow higher than receive seqcount on
1484 * clear because the rotation performed by clear will push
1485 * the index_sent_seqcount ahead (see
1486 * viewer_stream_sync_tracefile_array_tail) and skip over
1487 * packet sequence numbers.
1489 index
->status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1490 DBG("Check index status: retry for stream %" PRIu64
,
1491 vstream
->stream
->stream_handle
);
1493 } else if (!tracefile_array_seq_in_file(rstream
->tfa
,
1494 vstream
->current_tracefile_id
,
1495 vstream
->index_sent_seqcount
)) {
1497 * The next index we want to send cannot be read either
1498 * because we need to perform a rotation, or due to
1499 * the producer having overwritten its trace file.
1501 DBG("Viewer stream %" PRIu64
" rotation",
1502 vstream
->stream
->stream_handle
);
1503 ret
= viewer_stream_rotate(vstream
);
1505 /* EOF across entire stream. */
1506 index
->status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1510 * If we have been pushed due to overwrite, it
1511 * necessarily means there is data that can be read in
1512 * the stream. If we rotated because we reached the end
1513 * of a tracefile, it means the following tracefile
1514 * needs to contain at least one index, else we would
1515 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1516 * viewer. The updated index_sent_seqcount needs to
1517 * point to a readable index entry now.
1519 * In the case where we "rotate" on a single file, we
1520 * can end up in a case where the requested index is
1521 * still unavailable.
1523 if (rstream
->tracefile_count
== 1 &&
1524 !tracefile_array_seq_in_file(
1526 vstream
->current_tracefile_id
,
1527 vstream
->index_sent_seqcount
)) {
1528 index
->status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1529 DBG("Check index status: retry: "
1530 "tracefile array sequence number %" PRIu64
1531 " not in file for stream %" PRIu64
,
1532 vstream
->index_sent_seqcount
,
1533 vstream
->stream
->stream_handle
);
1536 LTTNG_ASSERT(tracefile_array_seq_in_file(rstream
->tfa
,
1537 vstream
->current_tracefile_id
,
1538 vstream
->index_sent_seqcount
));
1540 /* ret == 0 means successful so we continue. */
1545 viewer_stream_put(vstream
);
1551 void viewer_stream_rotate_to_trace_chunk(struct relay_viewer_stream
*vstream
,
1552 struct lttng_trace_chunk
*new_trace_chunk
)
1554 lttng_trace_chunk_put(vstream
->stream_file
.trace_chunk
);
1556 if (new_trace_chunk
) {
1557 const bool acquired_reference
= lttng_trace_chunk_get(
1560 LTTNG_ASSERT(acquired_reference
);
1563 vstream
->stream_file
.trace_chunk
= new_trace_chunk
;
1564 viewer_stream_sync_tracefile_array_tail(vstream
);
1565 viewer_stream_close_files(vstream
);
1569 * Send the next index for a stream.
1571 * Return 0 on success or else a negative value.
1574 int viewer_get_next_index(struct relay_connection
*conn
)
1577 struct lttng_viewer_get_next_index request_index
;
1578 struct lttng_viewer_index viewer_index
;
1579 struct ctf_packet_index packet_index
;
1580 struct relay_viewer_stream
*vstream
= NULL
;
1581 struct relay_stream
*rstream
= NULL
;
1582 struct ctf_trace
*ctf_trace
= NULL
;
1583 struct relay_viewer_stream
*metadata_viewer_stream
= NULL
;
1587 DBG("Viewer get next index");
1589 memset(&viewer_index
, 0, sizeof(viewer_index
));
1590 health_code_update();
1592 ret
= recv_request(conn
->sock
, &request_index
, sizeof(request_index
));
1596 health_code_update();
1598 vstream
= viewer_stream_get_by_id(be64toh(request_index
.stream_id
));
1600 DBG("Client requested index of unknown stream id %" PRIu64
,
1601 (uint64_t) be64toh(request_index
.stream_id
));
1602 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1606 /* Use back. ref. Protected by refcounts. */
1607 rstream
= vstream
->stream
;
1608 ctf_trace
= rstream
->trace
;
1610 /* metadata_viewer_stream may be NULL. */
1611 metadata_viewer_stream
=
1612 ctf_trace_get_viewer_metadata_stream(ctf_trace
);
1614 pthread_mutex_lock(&rstream
->lock
);
1617 * The viewer should not ask for index on metadata stream.
1619 if (rstream
->is_metadata
) {
1620 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1624 if (rstream
->ongoing_rotation
.is_set
) {
1625 /* Rotation is ongoing, try again later. */
1626 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1630 if (rstream
->trace
->session
->ongoing_rotation
) {
1631 /* Rotation is ongoing, try again later. */
1632 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1637 * Transition the viewer session into the newest trace chunk available.
1639 if (!lttng_trace_chunk_ids_equal(
1640 conn
->viewer_session
->current_trace_chunk
,
1641 rstream
->trace_chunk
)) {
1642 DBG("Relay stream and viewer chunk ids differ");
1644 ret
= viewer_session_set_trace_chunk_copy(
1645 conn
->viewer_session
,
1646 rstream
->trace_chunk
);
1648 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1654 * Transition the viewer stream into the latest trace chunk available.
1656 * Note that the stream must _not_ rotate in one precise condition:
1657 * the relay stream has rotated to a NULL trace chunk and the viewer
1658 * stream is consuming the trace chunk that was active just before
1659 * that rotation to NULL.
1661 * This allows clients to consume all the packets of a trace chunk
1662 * after a session's destruction.
1664 if (conn
->viewer_session
->current_trace_chunk
!= vstream
->stream_file
.trace_chunk
&&
1665 !(rstream
->completed_rotation_count
== vstream
->last_seen_rotation_count
+ 1 && !rstream
->trace_chunk
)) {
1666 DBG("Viewer session and viewer stream chunk differ: "
1667 "vsession chunk %p vstream chunk %p",
1668 conn
->viewer_session
->current_trace_chunk
,
1669 vstream
->stream_file
.trace_chunk
);
1670 viewer_stream_rotate_to_trace_chunk(vstream
,
1671 conn
->viewer_session
->current_trace_chunk
);
1672 vstream
->last_seen_rotation_count
=
1673 rstream
->completed_rotation_count
;
1676 ret
= check_index_status(vstream
, rstream
, ctf_trace
, &viewer_index
);
1679 } else if (ret
== 1) {
1681 * We have no index to send and check_index_status has populated
1682 * viewer_index's status.
1686 /* At this point, ret is 0 thus we will be able to read the index. */
1689 /* Try to open an index if one is needed for that stream. */
1690 ret
= try_open_index(vstream
, rstream
);
1691 if (ret
== -ENOENT
) {
1692 if (rstream
->closed
) {
1693 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1696 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_RETRY
);
1701 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1706 * vstream->stream_fd may be NULL if it has been closed by
1707 * tracefile rotation, or if we are at the beginning of the
1708 * stream. We open the data stream file here to protect against
1709 * overwrite caused by tracefile rotation (in association with
1710 * unlink performed before overwrite).
1712 if (!vstream
->stream_file
.handle
) {
1713 char file_path
[LTTNG_PATH_MAX
];
1714 enum lttng_trace_chunk_status status
;
1715 struct fs_handle
*fs_handle
;
1717 ret
= utils_stream_file_path(rstream
->path_name
,
1718 rstream
->channel_name
, rstream
->tracefile_size
,
1719 vstream
->current_tracefile_id
, NULL
, file_path
,
1726 * It is possible the the file we are trying to open is
1727 * missing if the stream has been closed (application exits with
1728 * per-pid buffers) and a clear command has been performed.
1730 status
= lttng_trace_chunk_open_fs_handle(
1731 vstream
->stream_file
.trace_chunk
,
1732 file_path
, O_RDONLY
, 0, &fs_handle
, true);
1733 if (status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1734 if (status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
&&
1736 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_HUP
);
1739 PERROR("Failed to open trace file for viewer stream");
1742 vstream
->stream_file
.handle
= fs_handle
;
1745 ret
= check_new_streams(conn
);
1747 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1749 } else if (ret
== 1) {
1750 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_STREAM
;
1753 ret
= lttng_index_file_read(vstream
->index_file
, &packet_index
);
1755 ERR("Relay error reading index file");
1756 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_ERR
);
1759 viewer_index
.status
= htobe32(LTTNG_VIEWER_INDEX_OK
);
1760 vstream
->index_sent_seqcount
++;
1764 * Indexes are stored in big endian, no need to switch before sending.
1766 DBG("Sending viewer index for stream %" PRIu64
" offset %" PRIu64
,
1767 rstream
->stream_handle
,
1768 (uint64_t) be64toh(packet_index
.offset
));
1769 viewer_index
.offset
= packet_index
.offset
;
1770 viewer_index
.packet_size
= packet_index
.packet_size
;
1771 viewer_index
.content_size
= packet_index
.content_size
;
1772 viewer_index
.timestamp_begin
= packet_index
.timestamp_begin
;
1773 viewer_index
.timestamp_end
= packet_index
.timestamp_end
;
1774 viewer_index
.events_discarded
= packet_index
.events_discarded
;
1775 viewer_index
.stream_id
= packet_index
.stream_id
;
1779 pthread_mutex_unlock(&rstream
->lock
);
1782 if (metadata_viewer_stream
) {
1783 pthread_mutex_lock(&metadata_viewer_stream
->stream
->lock
);
1784 DBG("get next index metadata check: recv %" PRIu64
1786 metadata_viewer_stream
->stream
->metadata_received
,
1787 metadata_viewer_stream
->metadata_sent
);
1788 if (!metadata_viewer_stream
->stream
->metadata_received
||
1789 metadata_viewer_stream
->stream
->metadata_received
>
1790 metadata_viewer_stream
->metadata_sent
) {
1791 viewer_index
.flags
|= LTTNG_VIEWER_FLAG_NEW_METADATA
;
1793 pthread_mutex_unlock(&metadata_viewer_stream
->stream
->lock
);
1796 viewer_index
.flags
= htobe32(viewer_index
.flags
);
1797 health_code_update();
1799 ret
= send_response(conn
->sock
, &viewer_index
, sizeof(viewer_index
));
1803 health_code_update();
1806 DBG("Index %" PRIu64
" for stream %" PRIu64
" sent",
1807 vstream
->index_sent_seqcount
,
1808 vstream
->stream
->stream_handle
);
1811 if (metadata_viewer_stream
) {
1812 viewer_stream_put(metadata_viewer_stream
);
1815 viewer_stream_put(vstream
);
1820 pthread_mutex_unlock(&rstream
->lock
);
1821 if (metadata_viewer_stream
) {
1822 viewer_stream_put(metadata_viewer_stream
);
1824 viewer_stream_put(vstream
);
1829 * Send the next index for a stream
1831 * Return 0 on success or else a negative value.
1834 int viewer_get_packet(struct relay_connection
*conn
)
1839 struct lttng_viewer_get_packet get_packet_info
;
1840 struct lttng_viewer_trace_packet reply_header
;
1841 struct relay_viewer_stream
*vstream
= NULL
;
1842 uint32_t reply_size
= sizeof(reply_header
);
1843 uint32_t packet_data_len
= 0;
1847 DBG2("Relay get data packet");
1849 health_code_update();
1851 ret
= recv_request(conn
->sock
, &get_packet_info
,
1852 sizeof(get_packet_info
));
1856 health_code_update();
1858 /* From this point on, the error label can be reached. */
1859 memset(&reply_header
, 0, sizeof(reply_header
));
1860 stream_id
= (uint64_t) be64toh(get_packet_info
.stream_id
);
1862 vstream
= viewer_stream_get_by_id(stream_id
);
1864 DBG("Client requested packet of unknown stream id %" PRIu64
,
1866 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1867 goto send_reply_nolock
;
1869 packet_data_len
= be32toh(get_packet_info
.len
);
1870 reply_size
+= packet_data_len
;
1873 reply
= (char *) zmalloc(reply_size
);
1875 PERROR("packet reply zmalloc");
1876 reply_size
= sizeof(reply_header
);
1880 pthread_mutex_lock(&vstream
->stream
->lock
);
1881 lseek_ret
= fs_handle_seek(vstream
->stream_file
.handle
,
1882 be64toh(get_packet_info
.offset
), SEEK_SET
);
1883 if (lseek_ret
< 0) {
1884 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1885 " to offset %" PRIu64
,
1887 (uint64_t) be64toh(get_packet_info
.offset
));
1890 read_len
= fs_handle_read(vstream
->stream_file
.handle
,
1891 reply
+ sizeof(reply_header
), packet_data_len
);
1892 if (read_len
< packet_data_len
) {
1893 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1894 ", offset: %" PRIu64
,
1896 (uint64_t) be64toh(get_packet_info
.offset
));
1899 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_OK
);
1900 reply_header
.len
= htobe32(packet_data_len
);
1904 /* No payload to send on error. */
1905 reply_size
= sizeof(reply_header
);
1906 reply_header
.status
= htobe32(LTTNG_VIEWER_GET_PACKET_ERR
);
1910 pthread_mutex_unlock(&vstream
->stream
->lock
);
1914 health_code_update();
1917 memcpy(reply
, &reply_header
, sizeof(reply_header
));
1918 ret
= send_response(conn
->sock
, reply
, reply_size
);
1920 /* No reply to send. */
1921 ret
= send_response(conn
->sock
, &reply_header
,
1925 health_code_update();
1927 PERROR("sendmsg of packet data failed");
1931 DBG("Sent %u bytes for stream %" PRIu64
, reply_size
, stream_id
);
1937 viewer_stream_put(vstream
);
1943 * Send the session's metadata
1945 * Return 0 on success else a negative value.
1948 int viewer_get_metadata(struct relay_connection
*conn
)
1955 struct lttng_viewer_get_metadata request
;
1956 struct lttng_viewer_metadata_packet reply
;
1957 struct relay_viewer_stream
*vstream
= NULL
;
1961 DBG("Relay get metadata");
1963 health_code_update();
1965 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
1969 health_code_update();
1971 memset(&reply
, 0, sizeof(reply
));
1973 vstream
= viewer_stream_get_by_id(be64toh(request
.stream_id
));
1976 * The metadata stream can be closed by a CLOSE command
1977 * just before we attach. It can also be closed by
1978 * per-pid tracing during tracing. Therefore, it is
1979 * possible that we cannot find this viewer stream.
1980 * Reply back to the client with an error if we cannot
1983 DBG("Client requested metadata of unknown stream id %" PRIu64
,
1984 (uint64_t) be64toh(request
.stream_id
));
1985 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
1988 pthread_mutex_lock(&vstream
->stream
->lock
);
1989 if (!vstream
->stream
->is_metadata
) {
1990 ERR("Invalid metadata stream");
1994 if (vstream
->metadata_sent
>= vstream
->stream
->metadata_received
) {
1996 * The live viewers expect to receive a NO_NEW_METADATA
1997 * status before a stream disappears, otherwise they abort the
1998 * entire live connection when receiving an error status.
2000 * Clear feature resets the metadata_sent to 0 until the
2001 * same metadata is received again.
2003 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
2005 * The live viewer considers a closed 0 byte metadata stream as
2008 if (vstream
->metadata_sent
> 0) {
2009 vstream
->stream
->no_new_metadata_notified
= true;
2010 if (vstream
->stream
->closed
) {
2011 /* Release ownership for the viewer metadata stream. */
2012 viewer_stream_put(vstream
);
2018 if (vstream
->stream
->trace_chunk
&&
2019 !lttng_trace_chunk_ids_equal(
2020 conn
->viewer_session
->current_trace_chunk
,
2021 vstream
->stream
->trace_chunk
)) {
2022 /* A rotation has occurred on the relay stream. */
2023 DBG("Metadata relay stream and viewer chunk ids differ");
2025 ret
= viewer_session_set_trace_chunk_copy(
2026 conn
->viewer_session
,
2027 vstream
->stream
->trace_chunk
);
2029 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
2034 if (conn
->viewer_session
->current_trace_chunk
&&
2035 conn
->viewer_session
->current_trace_chunk
!=
2036 vstream
->stream_file
.trace_chunk
) {
2037 bool acquired_reference
;
2039 DBG("Viewer session and viewer stream chunk differ: "
2040 "vsession chunk %p vstream chunk %p",
2041 conn
->viewer_session
->current_trace_chunk
,
2042 vstream
->stream_file
.trace_chunk
);
2043 lttng_trace_chunk_put(vstream
->stream_file
.trace_chunk
);
2044 acquired_reference
= lttng_trace_chunk_get(conn
->viewer_session
->current_trace_chunk
);
2045 LTTNG_ASSERT(acquired_reference
);
2046 vstream
->stream_file
.trace_chunk
=
2047 conn
->viewer_session
->current_trace_chunk
;
2048 viewer_stream_close_files(vstream
);
2051 len
= vstream
->stream
->metadata_received
- vstream
->metadata_sent
;
2053 if (!vstream
->stream_file
.trace_chunk
) {
2054 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
2057 } else if (vstream
->stream_file
.trace_chunk
&&
2058 !vstream
->stream_file
.handle
&& len
> 0) {
2060 * Either this is the first time the metadata file is read, or a
2061 * rotation of the corresponding relay stream has occurred.
2063 struct fs_handle
*fs_handle
;
2064 char file_path
[LTTNG_PATH_MAX
];
2065 enum lttng_trace_chunk_status status
;
2066 struct relay_stream
*rstream
= vstream
->stream
;
2068 ret
= utils_stream_file_path(rstream
->path_name
,
2069 rstream
->channel_name
, rstream
->tracefile_size
,
2070 vstream
->current_tracefile_id
, NULL
, file_path
,
2077 * It is possible the the metadata file we are trying to open is
2078 * missing if the stream has been closed (application exits with
2079 * per-pid buffers) and a clear command has been performed.
2081 status
= lttng_trace_chunk_open_fs_handle(
2082 vstream
->stream_file
.trace_chunk
,
2083 file_path
, O_RDONLY
, 0, &fs_handle
, true);
2084 if (status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2085 if (status
== LTTNG_TRACE_CHUNK_STATUS_NO_FILE
) {
2086 reply
.status
= htobe32(LTTNG_VIEWER_NO_NEW_METADATA
);
2088 if (vstream
->stream
->closed
) {
2089 viewer_stream_put(vstream
);
2093 PERROR("Failed to open metadata file for viewer stream");
2096 vstream
->stream_file
.handle
= fs_handle
;
2098 if (vstream
->metadata_sent
!= 0) {
2100 * The client does not expect to receive any metadata
2101 * it has received and metadata files in successive
2102 * chunks must be a strict superset of one another.
2104 * Skip the first `metadata_sent` bytes to ensure
2105 * they are not sent a second time to the client.
2107 * Baring a block layer error or an internal error,
2108 * this seek should not fail as
2109 * `vstream->stream->metadata_received` is reset when
2110 * a relay stream is rotated. If this is reached, it is
2111 * safe to assume that
2112 * `metadata_received` > `metadata_sent`.
2114 const off_t seek_ret
= fs_handle_seek(fs_handle
,
2115 vstream
->metadata_sent
, SEEK_SET
);
2118 PERROR("Failed to seek metadata viewer stream file to `sent` position: pos = %" PRId64
,
2119 vstream
->metadata_sent
);
2120 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
2126 reply
.len
= htobe64(len
);
2127 data
= (char *) zmalloc(len
);
2129 PERROR("viewer metadata zmalloc");
2133 fd
= fs_handle_get_fd(vstream
->stream_file
.handle
);
2135 ERR("Failed to restore viewer stream file system handle");
2138 read_len
= lttng_read(fd
, data
, len
);
2139 fs_handle_put_fd(vstream
->stream_file
.handle
);
2141 if (read_len
< len
) {
2143 PERROR("Failed to read metadata file");
2147 * A clear has been performed which prevents the relay
2148 * from sending `len` bytes of metadata.
2150 * It is important not to send any metadata if we
2151 * couldn't read all the available metadata in one shot:
2152 * sending partial metadata can cause the client to
2153 * attempt to parse an incomplete (incoherent) metadata
2154 * stream, which would result in an error.
2156 const off_t seek_ret
= fs_handle_seek(
2157 vstream
->stream_file
.handle
, -read_len
,
2160 DBG("Failed to read metadata: requested = %" PRIu64
", got = %zd",
2165 PERROR("Failed to restore metadata file position after partial read");
2171 vstream
->metadata_sent
+= read_len
;
2172 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_OK
);
2177 reply
.status
= htobe32(LTTNG_VIEWER_METADATA_ERR
);
2180 health_code_update();
2182 pthread_mutex_unlock(&vstream
->stream
->lock
);
2184 ret
= send_response(conn
->sock
, &reply
, sizeof(reply
));
2188 health_code_update();
2191 ret
= send_response(conn
->sock
, data
, len
);
2197 DBG("Sent %" PRIu64
" bytes of metadata for stream %" PRIu64
, len
,
2198 (uint64_t) be64toh(request
.stream_id
));
2200 DBG("Metadata sent");
2206 viewer_stream_put(vstream
);
2212 * Create a viewer session.
2214 * Return 0 on success or else a negative value.
2217 int viewer_create_session(struct relay_connection
*conn
)
2220 struct lttng_viewer_create_session_response resp
;
2222 DBG("Viewer create session received");
2224 memset(&resp
, 0, sizeof(resp
));
2225 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_OK
);
2226 conn
->viewer_session
= viewer_session_create();
2227 if (!conn
->viewer_session
) {
2228 ERR("Allocation viewer session");
2229 resp
.status
= htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR
);
2234 health_code_update();
2235 ret
= send_response(conn
->sock
, &resp
, sizeof(resp
));
2239 health_code_update();
2247 * Detach a viewer session.
2249 * Return 0 on success or else a negative value.
2252 int viewer_detach_session(struct relay_connection
*conn
)
2255 struct lttng_viewer_detach_session_response response
;
2256 struct lttng_viewer_detach_session_request request
;
2257 struct relay_session
*session
= NULL
;
2258 uint64_t viewer_session_to_close
;
2260 DBG("Viewer detach session received");
2264 health_code_update();
2266 /* Receive the request from the connected client. */
2267 ret
= recv_request(conn
->sock
, &request
, sizeof(request
));
2271 viewer_session_to_close
= be64toh(request
.session_id
);
2273 if (!conn
->viewer_session
) {
2274 DBG("Client trying to detach before creating a live viewer session");
2275 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR
);
2279 health_code_update();
2281 memset(&response
, 0, sizeof(response
));
2282 DBG("Detaching from session ID %" PRIu64
, viewer_session_to_close
);
2284 session
= session_get_by_id(be64toh(request
.session_id
));
2286 DBG("Relay session %" PRIu64
" not found",
2287 (uint64_t) be64toh(request
.session_id
));
2288 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK
);
2292 ret
= viewer_session_is_attached(conn
->viewer_session
, session
);
2294 DBG("Not attached to this session");
2295 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR
);
2296 goto send_reply_put
;
2299 viewer_session_close_one_session(conn
->viewer_session
, session
);
2300 response
.status
= htobe32(LTTNG_VIEWER_DETACH_SESSION_OK
);
2301 DBG("Session %" PRIu64
" detached.", viewer_session_to_close
);
2304 session_put(session
);
2307 health_code_update();
2308 ret
= send_response(conn
->sock
, &response
, sizeof(response
));
2312 health_code_update();
2320 * live_relay_unknown_command: send -1 if received unknown command
2323 void live_relay_unknown_command(struct relay_connection
*conn
)
2325 struct lttcomm_relayd_generic_reply reply
;
2327 memset(&reply
, 0, sizeof(reply
));
2328 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2329 (void) send_response(conn
->sock
, &reply
, sizeof(reply
));
2333 * Process the commands received on the control socket
2336 int process_control(struct lttng_viewer_cmd
*recv_hdr
,
2337 struct relay_connection
*conn
)
2342 msg_value
= be32toh(recv_hdr
->cmd
);
2345 * Make sure we've done the version check before any command other then a
2346 * new client connection.
2348 if (msg_value
!= LTTNG_VIEWER_CONNECT
&& !conn
->version_check_done
) {
2349 ERR("Viewer conn value %" PRIu32
" before version check", msg_value
);
2354 switch (msg_value
) {
2355 case LTTNG_VIEWER_CONNECT
:
2356 ret
= viewer_connect(conn
);
2358 case LTTNG_VIEWER_LIST_SESSIONS
:
2359 ret
= viewer_list_sessions(conn
);
2361 case LTTNG_VIEWER_ATTACH_SESSION
:
2362 ret
= viewer_attach_session(conn
);
2364 case LTTNG_VIEWER_GET_NEXT_INDEX
:
2365 ret
= viewer_get_next_index(conn
);
2367 case LTTNG_VIEWER_GET_PACKET
:
2368 ret
= viewer_get_packet(conn
);
2370 case LTTNG_VIEWER_GET_METADATA
:
2371 ret
= viewer_get_metadata(conn
);
2373 case LTTNG_VIEWER_GET_NEW_STREAMS
:
2374 ret
= viewer_get_new_streams(conn
);
2376 case LTTNG_VIEWER_CREATE_SESSION
:
2377 ret
= viewer_create_session(conn
);
2379 case LTTNG_VIEWER_DETACH_SESSION
:
2380 ret
= viewer_detach_session(conn
);
2383 ERR("Received unknown viewer command (%u)",
2384 be32toh(recv_hdr
->cmd
));
2385 live_relay_unknown_command(conn
);
2395 void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
2399 (void) lttng_poll_del(events
, pollfd
);
2401 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
2402 fd_tracker_util_close_fd
, NULL
);
2404 ERR("Closing pollfd %d", pollfd
);
2409 * This thread does the actual work
2412 void *thread_worker(void *data
)
2416 struct lttng_poll_event events
;
2417 struct lttng_ht
*viewer_connections_ht
;
2418 struct lttng_ht_iter iter
;
2419 struct lttng_viewer_cmd recv_hdr
;
2420 struct relay_connection
*destroy_conn
;
2422 DBG("[thread] Live viewer relay worker started");
2424 rcu_register_thread();
2426 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LIVE_WORKER
);
2428 if (testpoint(relayd_thread_live_worker
)) {
2429 goto error_testpoint
;
2432 /* table of connections indexed on socket */
2433 viewer_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2434 if (!viewer_connections_ht
) {
2435 goto viewer_connections_ht_error
;
2438 ret
= create_named_thread_poll_set(&events
, 2,
2439 "Live viewer worker thread epoll");
2441 goto error_poll_create
;
2444 ret
= lttng_poll_add(&events
, live_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
2453 health_code_update();
2455 /* Infinite blocking call, waiting for transmission */
2456 DBG3("Relayd live viewer worker thread polling...");
2457 health_poll_entry();
2458 ret
= lttng_poll_wait(&events
, -1);
2462 * Restart interrupted system call.
2464 if (errno
== EINTR
) {
2473 * Process control. The control connection is prioritised so we don't
2474 * starve it with high throughput tracing data on the data
2477 for (i
= 0; i
< nb_fd
; i
++) {
2478 /* Fetch once the poll data */
2479 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
2480 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2482 health_code_update();
2484 /* Thread quit pipe has been closed. Killing thread. */
2485 ret
= check_thread_quit_pipe(pollfd
, revents
);
2491 /* Inspect the relay conn pipe for new connection. */
2492 if (pollfd
== live_conn_pipe
[0]) {
2493 if (revents
& LPOLLIN
) {
2494 struct relay_connection
*conn
;
2496 ret
= lttng_read(live_conn_pipe
[0],
2497 &conn
, sizeof(conn
));
2501 ret
= lttng_poll_add(&events
,
2503 LPOLLIN
| LPOLLRDHUP
);
2505 ERR("Failed to add new live connection file descriptor to poll set");
2508 connection_ht_add(viewer_connections_ht
, conn
);
2509 DBG("Connection socket %d added to poll", conn
->sock
->fd
);
2510 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2511 ERR("Relay live pipe error");
2514 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2518 /* Connection activity. */
2519 struct relay_connection
*conn
;
2521 conn
= connection_get_by_sock(viewer_connections_ht
, pollfd
);
2526 if (revents
& LPOLLIN
) {
2527 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, &recv_hdr
,
2528 sizeof(recv_hdr
), 0);
2530 /* Connection closed. */
2531 cleanup_connection_pollfd(&events
, pollfd
);
2532 /* Put "create" ownership reference. */
2533 connection_put(conn
);
2534 DBG("Viewer control conn closed with %d", pollfd
);
2536 ret
= process_control(&recv_hdr
, conn
);
2538 /* Clear the session on error. */
2539 cleanup_connection_pollfd(&events
, pollfd
);
2540 /* Put "create" ownership reference. */
2541 connection_put(conn
);
2542 DBG("Viewer connection closed with %d", pollfd
);
2545 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2546 cleanup_connection_pollfd(&events
, pollfd
);
2547 /* Put "create" ownership reference. */
2548 connection_put(conn
);
2550 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2551 connection_put(conn
);
2554 /* Put local "get_by_sock" reference. */
2555 connection_put(conn
);
2562 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
2564 /* Cleanup remaining connection object. */
2566 cds_lfht_for_each_entry(viewer_connections_ht
->ht
, &iter
.iter
,
2569 health_code_update();
2570 connection_put(destroy_conn
);
2574 lttng_ht_destroy(viewer_connections_ht
);
2575 viewer_connections_ht_error
:
2576 /* Close relay conn pipes */
2577 (void) fd_tracker_util_pipe_close(the_fd_tracker
, live_conn_pipe
);
2579 DBG("Viewer worker thread exited with error");
2581 DBG("Viewer worker thread cleanup complete");
2585 ERR("Health error occurred in %s", __func__
);
2587 health_unregister(health_relayd
);
2588 if (lttng_relay_stop_threads()) {
2589 ERR("Error stopping threads");
2591 rcu_unregister_thread();
2596 * Create the relay command pipe to wake thread_manage_apps.
2597 * Closed in cleanup().
2599 static int create_conn_pipe(void)
2601 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
2602 "Live connection pipe", live_conn_pipe
);
2605 int relayd_live_join(void)
2607 int ret
, retval
= 0;
2610 ret
= pthread_join(live_listener_thread
, &status
);
2613 PERROR("pthread_join live listener");
2617 ret
= pthread_join(live_worker_thread
, &status
);
2620 PERROR("pthread_join live worker");
2624 ret
= pthread_join(live_dispatcher_thread
, &status
);
2627 PERROR("pthread_join live dispatcher");
2631 cleanup_relayd_live();
2639 int relayd_live_create(struct lttng_uri
*uri
)
2641 int ret
= 0, retval
= 0;
2647 goto exit_init_data
;
2651 /* Check if daemon is UID = 0 */
2652 is_root
= !getuid();
2655 if (live_uri
->port
< 1024) {
2656 ERR("Need to be root to use ports < 1024");
2658 goto exit_init_data
;
2662 /* Setup the thread apps communication pipe. */
2663 if (create_conn_pipe()) {
2665 goto exit_init_data
;
2668 /* Init relay command queue. */
2669 cds_wfcq_init(&viewer_conn_queue
.head
, &viewer_conn_queue
.tail
);
2671 /* Set up max poll set size */
2672 if (lttng_poll_set_max_size()) {
2674 goto exit_init_data
;
2677 /* Setup the dispatcher thread */
2678 ret
= pthread_create(&live_dispatcher_thread
, default_pthread_attr(),
2679 thread_dispatcher
, (void *) NULL
);
2682 PERROR("pthread_create viewer dispatcher");
2684 goto exit_dispatcher_thread
;
2687 /* Setup the worker thread */
2688 ret
= pthread_create(&live_worker_thread
, default_pthread_attr(),
2689 thread_worker
, NULL
);
2692 PERROR("pthread_create viewer worker");
2694 goto exit_worker_thread
;
2697 /* Setup the listener thread */
2698 ret
= pthread_create(&live_listener_thread
, default_pthread_attr(),
2699 thread_listener
, (void *) NULL
);
2702 PERROR("pthread_create viewer listener");
2704 goto exit_listener_thread
;
2708 * All OK, started all threads.
2713 * Join on the live_listener_thread should anything be added after
2714 * the live_listener thread's creation.
2717 exit_listener_thread
:
2719 ret
= pthread_join(live_worker_thread
, &status
);
2722 PERROR("pthread_join live worker");
2727 ret
= pthread_join(live_dispatcher_thread
, &status
);
2730 PERROR("pthread_join live dispatcher");
2733 exit_dispatcher_thread
:
2736 cleanup_relayd_live();